Overview
This project was built while taking a course in digital image processing. I designed a pipeline that uses an LLM to interpret a driver’s surroundings from dashcam footage and alert the user of a potential accident, like a pedestrian walking out into the road or a car drifting into the driver’s lane.
What I built
- Wrote a Python script to scrape dashcam footage from Reddit. This included near-miss and near-accident clips, and was used as test data for the pipeline.
- Built a pipeline that feeds dashcam frames to a large language model (LLM) to interpret what’s happening in the scene.
- Ran the pipeline on over 40 videos to measure real-time performance and how reliably it identified potential collisions.
Technologies
Python, PyTorch, OpenCV, Hugging Face, LLMs, Real-Time Detection Transformers (RT-DETRs), YOLO CV Models